About
Personal InfoSec Blog:
https://danielbohannon.com
Personal…
https://danielbohannon.com
Personal…
Activity
5K followers
-
Daniel Bohannon reposted thisDaniel Bohannon reposted thisToday I'm launching the Private Beta (i.e. waitlist) for Branchmore, a tool to help engineers measure and improve how they use AI coding tools. I've been working on it for the past few months. It grew out of my own experiences building with AI, coaching engineers, and creating tools for teams. Two questions kept bugging me: how do we accelerate with AI? And as the token bills piled up, how do we streamline without losing speed? I created Branchmore to help me track and optimize AI usage based on my engineering community's feedback. In the last 3 months, I paid $573 for a mix of Claude, Codex, Gemini, and Cursor. That $573 got me $3.45k worth of tokens at unsubsidized API prices (4.5 billion of them). Given that I'm using ~25% of capacity across tools, I could have gotten at least ~$13.8k of tokens over those 3 months. Pretty good for $573 (6x value for what I actually used, ~24x if I'd maxed out my capacity). What if I wanted to save money instead of tokenmaxxing? I spoke to a number of engineers who were more interested in making the most of their $20/mo subscription. There are tools in the app to help you look for waste. So what is Branchmore? It's not just costs. As I've built it and spoken to engineers, Branchmore has grown to cover several dimensions, all aimed at helping you stay on top of your tooling and keep your personal code factory humming. Branchmore is a feedback loop for engineers accelerating their coding with AI. I think of it as continuous improvement between people and their agents. I'm building for individual engineers: boosting your productivity and giving them safety rails when you go on autopilot. I've been running a solid production Alpha with my first batch of users for a few months, and it's been eye-opening to see what you can improve when you reflect on your AI interactions. I'm looking for folks excited by AI who want to share feedback and help shape the product. This is a Private Beta, not a full launch just yet. It's still small and the features are rough around the edges. It's a quick 5-minute setup, and then you can return to coding as the Branchmore engine analyzes in the background! 🔗 Want in? Drop a comment or DM me. Sign up at Branchmore[dot]com.
-
Daniel Bohannon reposted thisMany are familiar with and have followed the Advanced Practices team. Whether the team was under Mandiant, FireEye, Mandiant again or Google Cloud, the AP team has provided attribution and analytical support to Mandiant Consulting Incident Response and Mandiant Threat Defense SOC investigations for over fifteen years. That name has served us well, but it is time for a change. As the threat landscape evolves, so has the nature of our work. To better reflect the value the team provides to the industry, we have retired the Advanced Practices name and are now Frontline Intelligence Operations (FIO). Every word in the new name is deliberate: * Frontline: Because that is where the truth is, and that's where our focus will always remain * Intelligence: Because our job isn't just to report data, but to provide the industry the context, analysis, and 'so-what' behind what we see. * Operations: Because we want these insights to drive actionable decisions across the industry What does this mean for you? Outwardly, you may see artifacts that reflect the FIO branding. Inwardly, nothing changes about the quality of the insights industry has come to rely on.
-
Daniel Bohannon reposted thisDaniel Bohannon reposted thisI think I should be flattered! First ITW package that specifically tries to evade SandyClaw (unsuccessfully), Found by @kirkderpca (X handle). SandyClaw report in comments! https://npmjs[.]com/package/visa-response/v/1.4.4-beta
-
Daniel Bohannon reposted thisIn our previous research, we demonstrated how cross-prompt injection could transform Microsoft Copilot email summaries into a phishing surface (CVE-2026-26133). In this latest research, we explored what happens when that same trust-transfer problem moves beyond the inbox and into the browser. By embedding attacker-controlled content into ordinary web pages, we observed phishing links, spoofed security alerts, QR-code phishing pivots, and remote image fetching rendered directly within a trusted ChatGPT interface. Check it out here: https://lnkd.in/dpKbpDm6 #AI #AISecurity #LLMSecurity #CrossPromptInjection #Permiso #OpenAI #XPIA #ThreatResearch #P0Labs #ChatGPTDaniel Bohannon reposted thisThe latest research from Andi Ahmeti and the P0 Labs team introduces ChatGPhish, a browser-based prompt injection technique that turns any web page into a phishing payload inside ChatGPT. The setup: 1. A user visits a page and asks ChatGPT to summarize it. 2. Attacker-controlled content on that page gets passed into the model, and the response comes back with live phishing links, spoofed account alerts, auto-fetched tracking pixels, and inline QR codes. This is all rendered inside the trusted ChatGPT interface. A user doesn't have to open a malicious email or click on anything suspicious. They just have to visit a page and ask for a summary. This is different from our prior Copilot injection research in that the browser is the delivery primitive. Any page the user visits (GitHub READMEs, documentation, blog posts, marketing sites) can carry the payload. The larger problem is that attacker-controlled content rendered through an AI assistant inherits the trust of that assistant, and the users can't see the seam. Read the full research here: https://lnkd.in/e2PRnRQy
-
Daniel Bohannon shared thisAndi Ahmeti is at it again, continuing his AI-related research and highlighting real-world vulnerabilities in real-time! If you're not following his research and the awesome detection work that he and his team do at Permiso Security, then you're missing out. Great work as always, Andi!Daniel Bohannon shared thisThe latest research from Andi Ahmeti and the P0 Labs team introduces ChatGPhish, a browser-based prompt injection technique that turns any web page into a phishing payload inside ChatGPT. The setup: 1. A user visits a page and asks ChatGPT to summarize it. 2. Attacker-controlled content on that page gets passed into the model, and the response comes back with live phishing links, spoofed account alerts, auto-fetched tracking pixels, and inline QR codes. This is all rendered inside the trusted ChatGPT interface. A user doesn't have to open a malicious email or click on anything suspicious. They just have to visit a page and ask for a summary. This is different from our prior Copilot injection research in that the browser is the delivery primitive. Any page the user visits (GitHub READMEs, documentation, blog posts, marketing sites) can carry the payload. The larger problem is that attacker-controlled content rendered through an AI assistant inherits the trust of that assistant, and the users can't see the seam. Read the full research here: https://lnkd.in/e2PRnRQy
-
Daniel Bohannon shared thisLooking forward to meeting up with all the security friends in Seattle! Happy to chat about cloud and coffee in equal proportions 😎 See you there!Daniel Bohannon shared thisThe Permiso team is happy to sponsor #fwdcloudsec again this year! We'll be in attendance next week in Seattle. If you're there, stop by booth 13 and chat with our Principal Security Researcher, Daniel Bohannon, and our VP of Marketing, Jared Elder!
-
Daniel Bohannon reposted thisDaniel Bohannon reposted thisWe're grateful for the coverage from Michael Vizard at Security Boulevard on what we announced last week. AI agents are both the fastest-growing identity type in most cloud environments and they're the least visible. They spawn sub-agents, call external tools and MCP servers, inherit credentials, and operate at machine speed with little oversight. We extended the Permiso platform to discover every managed and unmanaged AI agent, track runtime activity across the full agent lifecycle, and assist you in enforcing least privilege policies. Read Mike's full piece here: https://hubs.la/Q04g-hlY0Permiso Security Adds Ability to Discover and Secure AI AgentsPermiso Security Adds Ability to Discover and Secure AI Agents
-
Daniel Bohannon reposted thisDaniel Bohannon reposted thisEveryone is racing to become an AI security company. We didn't have to. For five years, we've been building a platform that sees every identity, human and non-human, across cloud, SaaS, IdPs, and CI/CD, at runtime. When agentic AI showed up, it didn't break our model. It validated it. An AI agent is an identity. It authenticates, calls tools, accesses data, and spawns sub-agents that inherit permissions nobody reviewed. The security problem isn't new (over permissioned identities, poor AuthN/AuthZ controls, stale/zombie accounts, limited cross-environment inventory or visibility). The speed and autonomy are. Today, we shipped AI Agent Runtime Security across the Permiso platform. Every agent action - runs, tool calls, MCP invocations, data access - tied back to a specific identity in real time. Not during the forensics investigation. While it's happening. We didn't acquire our way here. We didn't slap .ai on our domain. We didn't pivot. We extended the platform our customers already trust to a new identity class that was always coming. Link to my blog post in the comments.
-
Daniel Bohannon reposted thisIt’s great when you don’t have to re-invent yourself or slap .ai on your domain name. Agentic AI is an identity (a composite one at that) and we are a run-time identity powerhouse. It was a natural extension for us.Daniel Bohannon reposted thisToday we're announcing AI agent runtime security across the Permiso platform. Permiso delivers identity runtime attribution for AI agents. Every run, every tool call, every MCP invocation tied to a specific human, non-human, or AI identity, with anomaly detection and least-privilege recommendations built on the same Universal Identity Graph and P0 Labs threat intelligence customers already use Sebastian Goodwin, Chief Trust Officer at Autodesk: "Permiso was already our security platform for identities. The natural next step was to partner with them for agentic AI identities." Discover, Protect, and Defend, now extended to the fastest-growing and least-governed identity class in the enterprise. Founder Jason Martin's Blog: https://hubs.la/Q04gyP-20
-
Daniel Bohannon liked thisDaniel Bohannon liked thisToday I'm launching the Private Beta (i.e. waitlist) for Branchmore, a tool to help engineers measure and improve how they use AI coding tools. I've been working on it for the past few months. It grew out of my own experiences building with AI, coaching engineers, and creating tools for teams. Two questions kept bugging me: how do we accelerate with AI? And as the token bills piled up, how do we streamline without losing speed? I created Branchmore to help me track and optimize AI usage based on my engineering community's feedback. In the last 3 months, I paid $573 for a mix of Claude, Codex, Gemini, and Cursor. That $573 got me $3.45k worth of tokens at unsubsidized API prices (4.5 billion of them). Given that I'm using ~25% of capacity across tools, I could have gotten at least ~$13.8k of tokens over those 3 months. Pretty good for $573 (6x value for what I actually used, ~24x if I'd maxed out my capacity). What if I wanted to save money instead of tokenmaxxing? I spoke to a number of engineers who were more interested in making the most of their $20/mo subscription. There are tools in the app to help you look for waste. So what is Branchmore? It's not just costs. As I've built it and spoken to engineers, Branchmore has grown to cover several dimensions, all aimed at helping you stay on top of your tooling and keep your personal code factory humming. Branchmore is a feedback loop for engineers accelerating their coding with AI. I think of it as continuous improvement between people and their agents. I'm building for individual engineers: boosting your productivity and giving them safety rails when you go on autopilot. I've been running a solid production Alpha with my first batch of users for a few months, and it's been eye-opening to see what you can improve when you reflect on your AI interactions. I'm looking for folks excited by AI who want to share feedback and help shape the product. This is a Private Beta, not a full launch just yet. It's still small and the features are rough around the edges. It's a quick 5-minute setup, and then you can return to coding as the Branchmore engine analyzes in the background! 🔗 Want in? Drop a comment or DM me. Sign up at Branchmore[dot]com.
-
Daniel Bohannon reacted on thisMany are familiar with and have followed the Advanced Practices team. Whether the team was under Mandiant, FireEye, Mandiant again or Google Cloud, the AP team has provided attribution and analytical support to Mandiant Consulting Incident Response and Mandiant Threat Defense SOC investigations for over fifteen years. That name has served us well, but it is time for a change. As the threat landscape evolves, so has the nature of our work. To better reflect the value the team provides to the industry, we have retired the Advanced Practices name and are now Frontline Intelligence Operations (FIO). Every word in the new name is deliberate: * Frontline: Because that is where the truth is, and that's where our focus will always remain * Intelligence: Because our job isn't just to report data, but to provide the industry the context, analysis, and 'so-what' behind what we see. * Operations: Because we want these insights to drive actionable decisions across the industry What does this mean for you? Outwardly, you may see artifacts that reflect the FIO branding. Inwardly, nothing changes about the quality of the insights industry has come to rely on.
-
Daniel Bohannon liked thisDaniel Bohannon liked thisЗа последние несколько лет через моё менторство прошло около 10 специалистов в области информационной безопасности и IT. Кто-то приходил с базовыми знаниями, кто-то уже работал в профессии, но не понимал, как перейти на следующий уровень. Мы разбирали реальные кейсы, архитектуру, управление рисками, безопасность разработки, инфраструктуру, проходили собеседования, учились презентовать свои решения бизнесу и брать на себя ответственность за результат. Самое приятное — наблюдать за результатами. Сегодня многие из моих бывших падаванов работают на позициях Senior Engineer, Security Lead, Security Architect и руководителей направлений. У некоторых доход вырос в 1,5 раза, у некоторых — в 2 раза и более. Но дело никогда не было только в деньгах. Для меня главный показатель успеха — когда человек перестаёт искать готовые ответы и начинает самостоятельно принимать сложные решения. Когда вчерашний ученик приходит уже не за советом, а обсудить архитектуру новой системы, стратегию безопасности или развитие команды. В такие моменты понимаешь, что знания — это единственный актив, который становится больше, когда им делишься. И, пожалуй, одна из самых приятных частей профессии — видеть, как твои падаваны становятся мастерами. #Менторство #Карьера #ИнформационнаяБезопасность #CyberSecurity #Leadership #Mentoring #SecurityArchitecture #CareerGrowth #KnowledgeSharing #DevSecOps #CISO
-
Daniel Bohannon liked thisDaniel Bohannon liked thisSpecial Token Injection (STI) Attack Guide @ BSides Porto 🛸🛸🛸🛸 https://lnkd.in/gKq-MQfn
-
Daniel Bohannon liked thisDaniel Bohannon liked thisIf an attacker disables GCP audit logging across your entire organization, your SIEM gets the event, but it can't tell you what happened. Art Ukshini from our P0 Labs team found that a GCP audit log field called serviceData (still actively used by multiple services) behaves inconsistently when exported to downstream analytics platforms. For specific high-value events, the field arrives stripped of its payload. Enabling and disabling audit logging will then produce indistinguishable log records. Detections built against this field will fail silently as there are no errors or alerts. The system appears healthy while the coverage is gone. What makes this worse is that the events most affected are the ones that matter most. Disabling cloud audit logs is a documented defense evasion tactic, and it's exactly the event that arrives incomplete. If you're writing detection logic against GCP logs outside the Chronicle ecosystem, read the full research here: https://hubs.la/Q04lBb6M0Mind the Gap: GCP serviceData in Logs Explorer vs. Exported LogsMind the Gap: GCP serviceData in Logs Explorer vs. Exported Logs
-
Daniel Bohannon liked thisDaniel Bohannon liked thisКиберқылмыскерлер қызық қулық тактиканы ойлап тапқан екен. Олар ChatGPT немесе Gemini сияқты ЖИ жүйелерінің этикалық шектеулерін (Guardrails) өздеріне қалқан етіп алған. Хакерлер ЖИ-ді екі түрлі жолмен алдап соғуда: 1️⃣ Вирусты ЖИ-мен қорғау (Код ішіндегі блокада): Хакерлер вирус кодының ішіне ЖИ ережелері бойынша қатаң тыйым салынған тақырыптарды (мысалы, қару, у, есірткі жасау мәтіндерін) әдейі қосып қояды. Киберқауіпсіздік маманы табылған вирусты ЖИ-ге салып, "Мына кодты талдап бер" десе, ЖИ код ішіндегі тыйым салынған сөздерді көріп, этикалық блок қояды да, талдаудан бас тартады. ЖИ өз ережелерімен хакердің қаруын зерттеуден қорғап қалады. 2️⃣ "Тұзақ" ортаны анықтау (Kill-switch): Вирус құрылғыға түскен соң, ЖИ-ге өрескел промпт жібереді. Егер ЖИ оған блок қойса, вирус "Мені зертханада (Sandbox) тексеріп жатыр" деп түсінеді де, өшіп қалады. Ал ЖИ қалыпты жауап берсе, вирус шабуылды жалғастырады.
Experience & Education
-
Permiso Security
View Daniel’s full experience
See their title, tenure and more.
or
Licenses & Certifications
-
ITIL Foundation
ITIL
IssuedCredential ID 9980050518886259 -
CISSP (Certified Information Systems Security Professional)
(ISC)²
Issued ExpiresCredential ID 518923 -
-
Languages
-
American Sign Language
Limited working proficiency
-
Turkish
Elementary proficiency
-
Albanian
Elementary proficiency
-
Kazakh
Elementary proficiency
Other similar profiles
Explore more posts
-
Karl Steinkamp
🔐 PCI DSS v4.0.1 – Clearing Up a Common Misunderstanding One of the most misunderstood areas of PCI DSS v4.0.1 is the use of customized approaches vs. compensating controls. Many organizations assume they’re interchangeable. They’re not. ➡️ Customized Approach = A designed control that meets the objective of a requirement in a different way than prescribed. It must be risk-based, documented, and validated by the assessor. ➡️ Compensating Control = A temporary measure used only when a requirement cannot be met as stated, and only if it provides equal or greater security. The difference matters. Why? - Misclassifying can lead to compliance gaps. - Customized approaches strengthen flexibility. - Compensating controls highlight areas that need remediation. 💡 PCI DSS v4.0.1 isn’t about “checking the box.” It’s about showing security intent and demonstrating how your controls meet the standard’s security objective. The Coalfire team can help your organization navigate PCI and many other regulatory, cybersecurity, and/or compliance needs.
1 Comment -
Michael Ransier
🚨 URGENT EXECUTIVE DIRECTIVE: Adversaries are actively leveraging zero-day and unauthenticated rce to compromise critical perimeter nodes. This executive brief breaks down the immediate operational risk, technical exposure, and remediation path for two high-impact vulnerabilities recently added to the CISA (KEV) catalog: CVE-2026-10520 and CVE-2026-35273. https://lnkd.in/gDxQFcA6
-
Simon Amoni
CISA has added an actively exploited high-severity vulnerability in OSGeo GeoServer to its KEV Catalog. Successful exploitation could allow attackers to read arbitrary files from the server’s file system through an XML External Entity (XXE) attack. Check out the blog for vulnerability details and how to fix it. #ThreatProtection #VulnerabilityManagement
-
Hadrian
Are you measuring the risks that really matter? Traditional assessments rarely capture the sequence of actions attackers use to exploit gaps. At RSAC 2026, we’ll explore how simulating real attacker behavior uncovers exposure hiding in plain sight. Using insights from the 2026 Offensive Security Benchmark Report, we’ll break down reconnaissance, attack paths, and exploitation steps that quantify real enterprise risk. Our CEO, Rogier Fischer, along with our team of experts Matan Shavit and Guy Stockwell, will be at Moscone Center, San Francisco, March 23‑26, sharing these insights with professionals tackling the same risks. If you want to map your blind spots before attackers do, book a time with our team at the conference: https://lnkd.in/e_4dMt2F #RSAC2026 #RSAC #Cybersecurityevents
1 Comment
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top contentAdd new skills with these courses
-
1h 38m
Ethical Hacking: Vulnerability Analysis
-
13m
A Standalone Project: Build a Program to Encrypt and Decrypt Text Messages Using an Encryption Algorithm to Protect Data from Unauthorized Access
-
13m
A Standalone Project: Build a Simple Intrusion Detection System (IDS) Using Python to Monitor Network Traffic